In the previous section we concluded that an efficient (accurate and fast) model-Based OPC target correction is needed to achieve better lithography-Friendly targets. This technique needs to be accurate and computationally efficient so that it can be effectively used in the OPC production environment. Also, if this approach can catch most of the retargeting needed to have a PW-friendly OPC target, then this would enable two important benefits. First, on the fabs side, reducing the computation needed for better yield. This is mainly because the post-AIR OPC target is more lithography-friendly and accordingly simpler OPC recipes are needed. Second, it is possible to model the PWOPC systematic deviation between the design intent and the final on-wafer characteristics.
Aerial Image Retargeting (AIR) identifies the weak PW designs (hot-spots) and classi- fies them into width sensitive and space-sensitive. The basic concept is based on evaluating the optical signature of the design (which also includes the proximity effects). The opti- cal signature (also known as the aerial image signature, because the resist effects are not included in it) is computed for the OPC target (i.e. its a pre-OPC simulation). Then the design is divided into small fragments (similar to what happens in regular OPC), but their movement (retargeting value) is extracted from a look-up table. This look-up table is developed by the DFM and OPC Engineers after training the recipe on a wide and yield-challenging design space. This movement is a change in the OPC target and should
not be confused with the OPC where the target fragments move to minimize their Edge Placement Error (EPE) between the printing image and the design intent.
Similar to RBRT, where the resizing value is determined by the width-space combi- nation, the AIR resizing value is coded in a table based on the Aerial Image signature using parameters like Imax, Imin (Maximum and Minimum optical intensity along the sim-
ulation site respectively), Aerial Image Slope, as well as the optical field curvature at the simulation site. This adds several degrees of freedom and is more capable of identifying Lithography hot-spots based on the Optics which is the root cause of Lithography limi- tations. Accordingly, even if the design is 2D it can be fixed independent of all different proximity effects and any surrounding designs.
Figures 4.1 and 4.2 show two different hot-spots (with the same width-space combi- nation), but in the same time they exhibit radically different post-OPC yield issues. It is obvious that both designs exhibit two very different aerial image signatures, which we claim to be the origin of the lithography yield issues. We propose using the aerial image signature in capturing and classifying lithography hot-spots because they can generically identify the lithography-related yield issues better than the geometrical description tech- niques. Moreover, using aerial image signatures in capturing hot-spots is much faster than relying on the full model-based lithography hot-spots capturing techniques (i.e. doing full OPC+lithography simulation).
To test the assumption above, we ran OPC on a chip that is designed to contain all the lithography-challenging designs. Then we mapped the PW worst widths (pinching) and spaces (bridging) to the aerial image maps. Figure 4.3 shows the AI map for the width- sensitive designs, where each contour represents hot-spots having the same measurement through the process-window, the smaller the number gets (on the legend) the worse it is from a lithography (yield) perspective. So for example in the figure the point (0.3,0.14) represents the worst design signature and results into severe pinching, then as the designs signatures starts to get away from this point, the failure severity starts to get better. Accordingly, we can conclude that for such map the designs with aerial image signature closest to (0.3,0.14) need a large positive bias (retargeting) to counter the effect of the bad initial design. One important observation from figure 4.3 is that the contours follow a monotonic behavior, which indicates that the AI signature can solely identify the PW weak areas. Moreover, using the aerial image contours curvature and the aerial image slope would add extra degrees of freedom in describing the PW more accurately and allows better separation between the both width and space weak designs signatures.
It is of particular interest to see if the aerial image contrast could alone serve as a single variable for identifying weak PW designs.
(a) Hot-Spot position and the critical direction
(b) Aerial Image distribution along the critical direc- tion
(c) Post-OPC through PW verification of the design showing bad soft pinching
(a) Hot-Spot position and the critical direction
(b) Aerial Image distribution along the critical direc- tion
(c) Post-OPC through PW verification of the design showing bad soft pinching
C = (Imax− Imin) (Imax+ Imin)
(4.1) In figure4.4, we are plotting the contours of constant contrast overlaid on the Imax-Imin
map shown earlier, while the color coding represents the severity of the hot-spot. It is very evident that the contrast cannot describe the weak PW areas by itself.
Figure 4.4: Mapping the contours of constant contrast on the Imax-Imin map (while the
severity level is color coded). It is shown that for the same contrast value passes through different regions of the map that goes between very critical and non critical.
AIR is thus capable of doing a Lithography-aware retargeting of the design that is very similar to that done during PWOPC (as PWOPC sacrifices the design fidelity (EPE) if the design is suffering from poor PW performance). However, AIR is capable of doing this retargeting 1) As a pre-OPC step, i.e. not linked to OPC, 2) Very computationally efficient (consumes less resources than that needed by a single nominal OPC iteration), 3) fits better in the parasitic-extraction flow and the modeling of the systematic PWOPC deviation from the original design as presented in figure7.28.
Figures 4.5(b)and4.5(b)show the standard mask tape-out flow and the proposed flow after the insertion of AIR as a MBRT module for lithography-related yield improvement. Note that with AIR, it is not necessary to use full PWOPC anymore and nominal-OPC is sufficient because the OPC target is more lithography friendly after MBRT. The AIR
correction flow is summarized in the flow chart in figure4.6 , where the design is simulated using the process optical model for all fragments, then looping over all fragments and comparing the aerial image signature to the AIR lookup table that defines the bias (re- targeting amount) needed for each AI signature, after than all fragments are allowed to move according to the pre-set values in the lookup table, followed by a cleanup step that ensures that no lithography target is being pushed into a different non-litho friendly regime. Figure 4.7 shows the methodology used for creating the AIR table, First a library of lithography challenging designs is used to evaluate the AI-PW performance correlation, then the generated table is used by the OPC engineer to set the proper fragments biasing based on their AI signature and then running the AIR mask tape-out flow and do PW OPC verification iteratively until the best AIR table testing is achieved. The last part of the flow is not easy to automate because of the importance of the human judgment, which makes building the AIR table a tedious job during the development but once the final table is achieved the impact of AIR is outstanding in terms of quality and runtime.